<p> As robotic technology moves out of factories and into broader segments of society, it promises to support a revolutionary improvement in the general standard of living. One of the largest hurdles to this increased use of robotic technology, however, is the inability of current mobile robots to negotiate difficult and delicate terrain in ways that are fast, efficient, and safe. Examples in nature demonstrate the incredible potential of legged locomotion to fill this need, but legged robots have not yet reached this level of performance. This work moves the field toward a better understanding of the design of robust and efficient legged robots by exploring the concept of reaction wheel actuation. This concept consists of the generation of torques on the robot's body via a high efficiency reaction wheel system (RWS), which operates by accelerating an internal reaction mass. These torques can be used to both improve the stability of the robot and increase its walking efficiency when used in a coordinated manner. </p><p> Due to the complex multi-body dynamics of these systems, the effect of design changes on a given performance metric are difficult to estimate. Full body trajectory optimization via direct transcription was employed as the primary tool to better understand the role that an RWS can play in bipedal locomotion. The five-link planar biped ERNIE was used as a basis for this analysis. Combined with a model of motor and gear losses, this allowed energetic comparisons between a baseline ERNIE model and hypothetical RWS-equipped versions. This showed for the first time that a correctly designed RWS, requiring only a modestly sized reaction wheel and a motor with minimal gear loss, can lead to improved walking efficiency.</p><p> Extensive optimization over the full operational walking range showed that a reasonably sized RWS with realistic levels of regenerative efficiency can improve walking efficiency by 5-10% over most gaits. Comparison of resulting gaits revealed that optimal RWS use leads to better centroidal momentum regulation, which in turn reduces loads on joint motors in the legs. Simulations of the ERNIE model under virtual constraint control verified these results. For validation, an experimental reaction wheel system was constructed and attached to the ERNIE biped. Walking experiments with ERNIE demonstrated a measurable improvement in walking efficiency for gaits that utilize the RWS, corroborating the results from both optimization and simulation. </p><p> For periodic walking, optimization results showed that near regions of marginal dynamic feasibility, reaction wheels can lead to much larger efficiency gains than in more typical operating conditions. RWS use also expands the range of dynamically feasible motions. In aperiodic motions such as speed and step length changes, RWS use is similarly beneficial, with significant efficiency gains in very demanding motions and expanded dynamic feasibility. For large, single-step changes of speed, the RWS can improve efficiency by as much as 60%. When transitions are executed over longer, multi-step duration, RWS use provides benefits similar to those seen in periodic walking. </p><p> The potential role of reaction wheels in improving balance was also examined. Design principles for efficient RWS stabilization were derived by combining RWS-based balance controllers with accurate gear and motor models. Stabilization generally demands more RWS torque than steady state walking, but still favors relatively light gearing in order to minimize gear loss and maximize momentum storage potential. A task space controller for underactuated balancing was developed to compare baseline and RWS-equipped bipeds in terms of balance performance. While the RWS-equipped biped was able to balance and reject large disturbances, the baseline biped was only able to balance for short periods of time.</p>

<p> In this dissertation, we present novel solutions to cooperative control of autonomous multi-agent network topologies pertaining to the area of hostile target tracking by multiple unmanned aerial vehicles (UAVs). The present work assumes an undirected graph comprising point-mass UAVs with time-varying communication topology among agents. The level of information sharing ability among agents in a multi-agent network, i.e. the <i>network connectivity,</i> plays pivotal role in group dynamics. A neighborhood information based decentralized controller is proposed in order to drive UAVs into a symmetric formation of polygon shape surrounding a mobile target, simultaneously with maintaining and controlling connectivity during the formation process. Appropriate controller parameter selection schemes, both for controller weights and gains, are adapted for dynamic topologies to maintain the connectivity measure above zero at all times. A challenging task of tracking a desired connectivity profile along with the formation control, is accomplished by using time-varying controller gains throughout agents dynamics. We next present a generalized formation controller, which in fact generates a family of UAV trajectories satisfying the control criteria. The proposed decentralized controller contains additional tuning parameters as fractional powers on proportional and derivative terms, rendering flexibility in achieving the control objective. The proposed controller with proper fractional powers, results in gradual state changes in UAV dynamics by using limited control inputs. Moreover, we extend our work by addressing a ground target tracking and reacquiring problem using the visual information gathered by flying UAV. The proposed guidance law uses line-of-sight guidance to track the target pushing it towards the image center captured by UAV, and exploits UAV-target mutual information to reacquire the target in case it steers away from the field-of-view for a short time. The convergence of the closed loop systems under the proposed controllers are shown using Lyapunov theory. Simulation results validate the effectiveness and novelty of the proposed control laws.</p><p> In addition to the above, this work focuses on categorizing multi-agent topologies in concern with the network dynamics and connectivity to analyze, realize, and visualize multi-agent interactions. In order to explore various useful agents reconfiguration possibilities without compromising the network connectivity, the present work aims at determining distinct topologies with the same connectivity or <i>isoconnected topologies.</i> Different topologies with identical connectivity are found out with the help of analytic techniques utilizing matrix algebra and calculus of variation. Elegant strategies for preserving connectivity in a network with a single mobile agent and rest of the stationary members, are proposed in this work as well. The proposed solutions are validated with the help of sufficient examples. For visual understanding of how agents locations and topology configurations influence the network connectivity, a MATLAB based graphical user interface is designed to interact with multi-agent graphs in a user-friendly manner.</p><p> To this end, the present work succeeds to determine solutions to challenging multi-UAV cooperative control problems, such as: (1) Symmetric formation control surrounding a mobile target; (2) Maintaining, improving and controlling the network connectivity during a mission; and (3) Categorizing different multi-agent topologies to unravel useful reconfiguration options for a group. The proposed theories with appropriate analysis, and the simulation results suffice to show the contribution and novelty of this work.</p>

<p>The project is in the area of autonomous navigation robots, with a focus on assisting the blind and physically disabled. The project addresses two main issues concerning the autonomous navigation. First, avoiding obstacles in robot?s path towards destination, solved by placing three ultrasonic sensors on the front, right, and left sides of the robot. They are effective in detecting obstacles in the range of 0.02 meters to 4 meters. Second, giving a path for the navigation, the robot travels through a series of waypoints to reach the destination. Heading angle and distance to each of these coordinates are calculated using Magnetometer and GPS. Using waypoints to reach destination gives user the flexibility to select a route and navigate without use of the internet. A Human Voice Interface system has been added, which enables the user to control basic robot actions through Bluetooth voice commands. The above model is effective in autonomously guiding the user to their destination. Results have been demonstrated and tested using a four-wheel robot model. LCD and serial monitor readings are presented with the paper. Efforts are being made to make this model more flexible so that it could be attached to any functional wheelchair with minor modifications.

Legged mobile robots have a number of distinct advantages over wheeled or flight-enabled platforms, including optimal foot placement and task- or terrain-specific gait strategies. When analyzing a legged robot, the use of reduced order models is common for understanding the platform dynamics and developing a controller template. In this context, if a single-legged platform can be optimized for a particular use (e.g. speed or efficiency, rough or smooth terrain, etc.), the resulting control scheme informs the control of a multi-legged platform, and is significantly easier to develop. However, the optimization of a gait controller on one robotic platform is platform-dependent: it is unclear from a single gait study whether the optimized parameters are effective only for a given terrain, mechanism, or physical scale. This thesis tests that platform dependence. In doing so, it orchestrates the tools for transposing a gait from one robotic system to another. It also reoptimizes a set of gaits on a new platform, elucidates terrain-specific and platform-specific behavior, and analyzes select gait features as essential or non-essential for energy-efficient running on a given terrain. In a comparison between two single-legged hopping robots, leg touchdown angle and leg angle modulation are consistent parameters across terrain, but vary between platforms. Aggressive PD terms increase energy-efficiency on deformable media, and appear to increase stability for excessive touchdown angles as well. Asymmetry in the gait of a 5-bar mechanism is an emergent property that appears to improve gait performance by injecting energy, though why this feature is necessary is unclear. Trajectories of successful gaits appear consistent between platforms and match biological inspiration (which was also an emergent property and not explicitly controlled). / A Thesis submitted to the Department of Mechanical Engineering in partial fulfillment of the requirements for the degree of Master of Science. / Summer Semester 2018. / July 17, 2018. / Includes bibliographical references. / Jonathan Clark, Professor Directing Thesis; Kourosh Shoele, Committee Member; Carl Moore, Committee Member.

Legged animals have long been shown to excel at maneuvering and navigating the complex and changing environments found in the natural world through a combination of highly actuated musculoskeletal structures, robust and flexible control schemes, and advanced sensory organs. Dynamic legged robots have been shown to generate some of this wide array of unique, high-energy locomotive behaviors (specifically the abilities to run, jump, and climb over and around obstacles) which makes them attractive candidates for robotic applications navigating the real world. However, although these robots have been designed to capture many of the behaviors of animals, current implementations can not match biological systems in terms of robustness, efficiency, and flexibility of motion. These limitations are exacerbated by the fact that the primary techniques used for robotic navigation are not currently equipped to utilize the full range of behaviors afforded to legged systems. This thesis addresses three aspects of dynamic legged locomotion necessary for field implementation, specifically with regards to running and climbing motions. The first thrust examines and directly compares two distinct, commonly-used controller archetypes in running and evaluates their the merits in terms of speed, efficiency, and robustness. The second thrust explores the effects of leg morphology on locomotive performance, motivating the construction of the design concept of "Effective Dynamic Workspace" which enabled dynamic running and climbing on a quadrupedal robot. Finally, the thesis concludes with the development of a motion planner which addresses the unique mobility constraints and enables navigation on a Full-Goldman style climbing robot. / A Thesis submitted to the Department of Mechanical Engineering in partial fulfillment of the requirements for the degree of Master of Science. / Fall Semester 2018. / November 8, 2018. / Includes bibliographical references. / Johnathan Clark, Professor Directing Thesis; Patrick Hollis, Committee Member; William Oates, Committee Member.

A fundamental challenge in deploying vision-based object detection on a robotic platform is achieving sufficient perceptual performance for safe and effective operation. While general purpose object detectors have steadily improved in performance, we are still some way from being able to rely solely on purely vision-based robotic perception systems. This thesis seeks to resolve this problem by exploiting a characteristic of computer vision unique to robotic applications, with a particular focus on pedestrian detection for autonomous driving. Robots - even mobile ones - typically have some inherent structure to the visual data they process. Firstly, they operate in the same environment (for example, regular commuting or errands with autonomous cars), and secondly, they are able to localise within that environment. Place matters in robotics, as we can exploit this additional context to boost perceptual performance by constructing object detector models fitted to a mobile robot's place of operation. We demonstrate that, in an ideal scenario with ground truth labels, we can significantly improve the detection performance of lightweight object detectors by exploiting place. Our results suggest that a key factor limiting detection performance is the model capacity, suggesting that this approach could equally be applied to higher capacity models as the computational budget dictates. This local expert detector is developed into a deployable, self-supervised system, using offline image segmentation and spatial heuristics to construct the detector models and a visual localisation system to retrieve them at run time. This approach boosts the perceptual performance of our lightweight object detector models. Finally, this ensemble approach to local expert object detection is extended further with a neural network trained to generate detector models conditioned on the input image, an approach we refer to as a Dynamic Detection Filter Network. The network learns a representation of the operating environment, generating model parameters based on the input image. This offers a general approach to constructing place specific object detectors independently of localisation, with the potential to operate on larger scales with many different environments.

Legged mobile robots have a number of distinct advantages over wheeled or flight-enabled platforms, including optimal foot placement and task- or terrain-specific gait strategies. When analyzing a legged robot, the use of reduced order models is common for understanding the platform dynamics and developing a controller template. In this context, if a single-legged platform can be optimized for a particular use (e.g. speed or efficiency, rough or smooth terrain, etc.), the resulting control scheme informs the control of a multi-legged platform, and is significantly easier to develop. However, the optimization of a gait controller on one robotic platform is platform-dependent: it is unclear from a single gait study whether the optimized parameters are effective only for a given terrain, mechanism, or physical scale. This thesis tests that platform dependence. In doing so, it orchestrates the tools for transposing a gait from one robotic system to another. It also reoptimizes a set of gaits on a new platform, elucidates terrain-specific and platform-specific behavior, and analyzes select gait features as essential or non-essential for energy-efficient running on a given terrain. In a comparison between two single-legged hopping robots, leg touchdown angle and leg angle modulation are consistent parameters across terrain, but vary between platforms. Aggressive PD terms increase energy-efficiency on deformable media, and appear to increase stability for excessive touchdown angles as well. Asymmetry in the gait of a 5-bar mechanism is an emergent property that appears to improve gait performance by injecting energy, though why this feature is necessary is unclear. Trajectories of successful gaits appear consistent between platforms and match biological inspiration (which was also an emergent property and not explicitly controlled). / A Thesis submitted to the Department of Mechanical Engineering in partial fulfillment of the requirements for the degree of Master of Science. / Summer Semester 2018. / July 17, 2018. / Includes bibliographical references. / Jonathan Clark, Professor Directing Thesis; Kourosh Shoele, Committee Member; Carl Moore, Committee Member.

This thesis describes contributions towards the implementation of Human-in-the-Loop (HitL) grasping for assistive robotics, with a particular focus on low throughput, high noise interfaces such as electroencephalography (EEG) or electromyography (EMG) brain-computer interface(BCI) devices in natural environments. Although progress in the robotics field has been swift, it is unlikely that truly independent operation of robots in situations where they will interact closely with objects, obstacles, and perhaps even other people in their environment will evolve in the immediate future. However, with the help of a human operator, it is possible to achieve robust, safe operation in complex environments. This work describes a system that can accomplish this with minimal interfaces that are accessible even to individuals with impairments, which will enable the development of more capable assistive devices for these individuals. Grasping an object generally requires contextual knowledge of the object and the intent of the user. We have developed a user interface for an on-line grasp planner that allows the user to effectively express their intent. Grasping in natural environments requires grasp planning algorithms that are robust to target localization errors. This work describes grasp quality measurements that generate more robust grasps by considering the local geometry of the object as well as how uncertainty will affect the proposed grasp. These new measures are integrated into an augmented reality interface that allows a user to plan a grasp online that matches their intent for using the object that is to be grasped. This interface is validated by testing with real users, both healthy and impaired, using a variety of input devices suitable for impaired subjects, such as low cost EEG and EMG devices. This work forms the foundation for a flexible, fully featured HitL system that will allow users to grasp objects in cluttered spaces using novel, practical BCI devices that have the potential to bring HiTL assistive devices out of the research environment and into the lives of those that need them.

Chow Man Kit. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2003. / Includes bibliographical references (leaves 88-92). / Abstracts in English and Chinese. / ABSTRACT --- p.i / 摘要 --- p.iii / ACKNOWLEDGEMENTS --- p.v / TABLE OF CONTENTS --- p.vi / LIST OF FIGURES --- p.ix / LIST OF TABLES --- p.xi / Chapter 1. --- INTRODUCTION --- p.1 / Chapter 1.1 --- Previous Models on Robot Software Architecture --- p.2 / Chapter 1.1.1 --- SPA --- p.2 / Chapter 1.1.2 --- Sub sumption Architecture --- p.3 / Chapter 1.1.3 --- Three Layer Architecture --- p.4 / Chapter 1.1.4 --- Two Layer Architecture --- p.6 / Chapter 1.1.5 --- RCS --- p.7 / Chapter 1.2 --- Motivation and Research Objective --- p.9 / Chapter 1.2.1 --- Motivation --- p.9 / Chapter 1.2.2 --- Contribution --- p.10 / Chapter 1.3 --- Thesis Outline --- p.11 / Chapter 2. --- STUDY ON ARCHITECTURE --- p.12 / Chapter 2.1 --- Hierarchy in Architecture --- p.12 / Chapter 2.1.1 --- Purpose of Hierarchy --- p.12 / Chapter 2.1.2 --- Suggested Hierarchy --- p.14 / Chapter 2.1.3 --- Short Summary in Hierarchy --- p.18 / Chapter 2.2 --- Modularity in Architecture --- p.18 / Chapter 2.2.1 --- Purpose of Modularity --- p.18 / Chapter 2.2.2 --- Suggested Modularity --- p.19 / Chapter 2.3 --- Connectivity in Architecture --- p.20 / Chapter 2.3.1 --- Purpose of Connectivity --- p.20 / Chapter 2.3.2 --- Suggested Connectivity --- p.21 / Chapter 3. --- STUDY ON INTERFACES --- p.23 / Chapter 3.1 --- Physical Interface --- p.24 / Chapter 3.2 --- Application Programming Interface (API) --- p.24 / Chapter 3.3 --- User Interface --- p.27 / Chapter 4. --- PROSPOSED ROBOT SOFTWARE ARCHITECTURE --- p.29 / Chapter 5. --- PRACTICAL IMPLEMENTATION --- p.32 / Chapter 5.1 --- Hardware Implementation --- p.32 / Chapter 5.1.1 --- Driving Module --- p.33 / Chapter 5.1.1.1 --- Wheels and motors arrangement --- p.36 / Chapter 5.1.1.2 --- Kinematics of wheeled mobile robot --- p.37 / Chapter 5.1.1.3 --- Inverse kinematics of the mobile robot --- p.41 / Chapter 5.1.1.4 --- Dynamic Controller --- p.44 / Chapter 5.1.1.5 --- Emergency Stop --- p.52 / Chapter 5.1.1.6 --- Homing Mechanism for Steering Axis --- p.53 / Chapter 5.1.2 --- Sensing Module --- p.54 / Chapter 5.1.2.1 --- Sensing System --- p.55 / Chapter 5.1.2.2 --- Using Comport as the Data Transmission Medium --- p.55 / Chapter 5.1.3 --- Power Configuration --- p.56 / Chapter 5.1.3.1 --- Basic Power Connection --- p.57 / Chapter 5.1.3.2 --- Design on Power Distribution System --- p.57 / Chapter 5.2 --- Software Considerations --- p.59 / Chapter 5.2.1 --- Operating System --- p.59 / Chapter 5.2.2 --- Parallel Processing --- p.59 / Chapter 5.3 --- Implementation of Robot Software Architecture --- p.61 / Chapter 5.3.1 --- Local Terminal Module --- p.62 / Chapter 5.3.2 --- Navigation Module --- p.62 / Chapter 5.3.3 --- Sensing Module --- p.64 / Chapter 5.3.3.1 --- Sensor Data Retrieval --- p.65 / Chapter 5.3.3.2 --- Error Checking --- p.65 / Chapter 5.3.3.3 --- Calculating Obstacle Repulsive Vector --- p.67 / Chapter 5.3.3.4 --- Visualizing Sensor Data --- p.67 / Chapter 5.3.4 --- Communication Module --- p.68 / Chapter 5.3.5 --- New idea integrated in Communication Module --- p.70 / Chapter 5.4 --- Summary --- p.73 / Chapter 6. --- APPICATION EXAMPLE AND EXPERIMENT --- p.76 / Chapter 6.1 --- Application Example --- p.77 / Chapter 6.2 --- Experiment --- p.78 / Chapter 7. --- CONCLUSIONS AND FUTURE WORKS --- p.81 / Chapter 7.1 --- Conclusions --- p.81 / Chapter 7.2 --- Future Works --- p.83 / APPENDIX --- p.84 / Chapter A. --- Homing Mechanism for Steering Axis --- p.84 / Chapter A.1 --- Working Algorithm --- p.84 / Chapter A.2 --- Hardware Component --- p.85 / Chapter A.3 --- Circuit Diagram --- p.85 / Chapter A.4 --- Pin Assignment --- p.85 / Chapter B. --- Power Specification --- p.86 / Chapter B.1 --- Power Consumption by PC --- p.86 / Chapter B.2 --- Hardware Component on Power --- p.87 / BIBLIOGRAPHY --- p.88

In order to fully exploit robot motion capabilities in complex environments, robots need to reason about obstacles in a non-binary fashion. In this paper, we focus on the modeling and characterization of pliable materials such as tall vegetation. These materials are of interest because they are pervasive in the real world, requiring the robotic vehicle to determine when to traverse or avoid them. This paper develops and experimentally verifies two template models for vegetation-rich terrain. In addition, it presents and validates a methodology to generate predictions of the associated energetic cost incurred by tracked and skid-steered mobile robots when traversing a vegetation patches of variable density. Another class of terrains considered in this work are regions of shallow, dense fluids, such as a beach-head, stream banks, snow or mud. This work examines the behavior of a simulated SLIP runner operating in such a viscous medium. Simulation results show that intelligently retracting the leg during flight can have a profound effect on the maximum achievable velocity of the runner, the stability of the resulting gait, and the cost of transport of the runner. Results also show that trudging gaits, in which the leg is positioned behind the center of mass, can be favorable in certain situations in terms of energy consumption and forward velocity. / A Thesis submitted to the Department of Mechanical Engineering in partial fulfillment of the requirements for the degree of Master of Science. / Spring Semester 2019. / April 16, 2019. / Fluid, Legged robot, Mobile robot, Rough Terrain, Running, Vegetation / Includes bibliographical references. / Jonathan Clark, Professor Directing Thesis; Christian Hubicki, Committee Member; Kourosh Shoele, Committee Member.